SQLSERVER排查CPU占用高的情況

今天中午，有朋友叫我幫他看一下數據庫，操作系統是Windows2008R2 ,數據庫是SQL2008R2 64位

64G內存，16核CPU

硬件配置還是比較高的，他說服務器運行的是金蝶K3軟件，數據庫實例里有多個數據庫

 

現象

他說是這幾天才出現的，而且在每天的某一個時間段才會出現CPU占用高的情況

內存占用不太高，只占用了30個G

CPU占用100%

---

排查方向

 

一般排查都是用下面的腳本，一般會用到三個視圖sys.sysprocesses ,dm_exec_sessions ,dm_exec_requests

USE master
GO
--如果要指定數據庫就把注釋去掉
SELECT * FROM sys.[sysprocesses] WHERE [spid]>50 --AND DB_NAME([dbid])='gposdb'
SELECT COUNT(*) FROM [sys].[dm_exec_sessions] WHERE [session_id]>50

看一下當前的數據庫用戶連接有多少

然后使用下面語句看一下各項指標是否正常，是否有阻塞，這個語句選取了前10個最耗CPU時間的會話

SELECT TOP 10
[session_id],
[request_id],
[start_time] AS '開始時間',
[status] AS '狀態',
[command] AS '命令',
dest.[text] AS 'sql語句', 
DB_NAME([database_id]) AS '數據庫名',
[blocking_session_id] AS '正在阻塞其他會話的會話ID',
[wait_type] AS '等待資源類型',
[wait_time] AS '等待時間',
[wait_resource] AS '等待的資源',
[reads] AS '物理讀次數',
[writes] AS '寫次數',
[logical_reads] AS '邏輯讀次數',
[row_count] AS '返回結果行數'
FROM sys.[dm_exec_requests] AS der 
CROSS APPLY 
sys.[dm_exec_sql_text](der.[sql_handle]) AS dest 
WHERE [session_id]>50 AND DB_NAME(der.[database_id])='gposdb'  
ORDER BY [cpu_time] DESC

如果想看具體的SQL語句可以執行下面的SQL語句，記得在SSMS里選擇以文本格式顯示結果

--在SSMS里選擇以文本格式顯示結果
SELECT TOP 10 
dest.[text] AS 'sql語句'
FROM sys.[dm_exec_requests] AS der 
CROSS APPLY 
sys.[dm_exec_sql_text](der.[sql_handle]) AS dest 
WHERE [session_id]>50  
ORDER BY [cpu_time] DESC

模擬了一些耗CPU時間的動作

 

還有查看CPU數和user scheduler數和最大工作線程數，檢查worker是否用完也可以排查CPU占用情況

--查看CPU數和user scheduler數目
SELECT cpu_count,scheduler_count FROM sys.dm_os_sys_info
--查看最大工作線程數
SELECT max_workers_count FROM sys.dm_os_sys_info

查看機器上的所有schedulers包括user 和system
通過下面語句可以看到worker是否用完，當達到最大線程數的時候就要檢查blocking了

對照下面這個表
各種CPU和SQLSERVER版本組合自動配置的最大工作線程數
CPU數                 32位計算機                        64位計算機
<=4                     256                                   512
  8                        288                                   576
 16                       352                                   704
 32                       480                                   960

SELECT
scheduler_address,
scheduler_id,
cpu_id,
status,
current_tasks_count,
current_workers_count,active_workers_count
FROM sys.dm_os_schedulers

 

如果SQLSERVER存在要等待的資源，那么執行下面語句就會顯示出會話中有多少個worker在等待

結合[sys].[dm_os_wait_stats]視圖，如果當前SQLSERVER里面沒有任何等待資源，那么下面的SQL語句不會顯示任何結果

SELECT TOP 10
 [session_id],
 [request_id],
 [start_time] AS '開始時間',
 [status] AS '狀態',
 [command] AS '命令',
 dest.[text] AS 'sql語句', 
 DB_NAME([database_id]) AS '數據庫名',
 [blocking_session_id] AS '正在阻塞其他會話的會話ID',
 der.[wait_type] AS '等待資源類型',
 [wait_time] AS '等待時間',
 [wait_resource] AS '等待的資源',
 [dows].[waiting_tasks_count] AS '當前正在進行等待的任務數',
 [reads] AS '物理讀次數',
 [writes] AS '寫次數',
 [logical_reads] AS '邏輯讀次數',
 [row_count] AS '返回結果行數'
 FROM sys.[dm_exec_requests] AS der 
 INNER JOIN [sys].[dm_os_wait_stats] AS dows 
 ON der.[wait_type]=[dows].[wait_type]
 CROSS APPLY 
 sys.[dm_exec_sql_text](der.[sql_handle]) AS dest 
 WHERE [session_id]>50  
 ORDER BY [cpu_time] DESC

比如我當前執行了查詢SalesOrderDetail_test表100次，由於表數據非常多，所以SSMS需要把SQLSERVER執行的結果慢慢的取走，

造成了ASYNC_NETWORK_IO等待

USE [AdventureWorks]
GO
SELECT * FROM dbo.[SalesOrderDetail_test]
GO 100

 

---

問題源頭

經過排查和這幾天的觀察情況，確定是某些表缺失索引導致，現在在這些表上增加了索引，問題解決

select * from t_AccessControl        --權限控制表權限控制
select * from t_GroupAccess            --用戶組權限表用戶組權限
select * from t_GroupAccessType        --用戶組權限類表用戶組權限類
select * from t_ObjectAccess        --對象權限表對象權限
select * from t_ObjectAccessType    --對象權限類型表對象權限類型
select * from t_ObjectType            --對象類型表對象類型

查詢CPU占用高的語句

SELECT TOP 10
   total_worker_time/execution_count AS avg_cpu_cost, plan_handle,
   execution_count,
   (SELECT SUBSTRING(text, statement_start_offset/2 + 1,
      (CASE WHEN statement_end_offset = -1
         THEN LEN(CONVERT(nvarchar(max), text)) * 2
         ELSE statement_end_offset
      END - statement_start_offset)/2)
   FROM sys.dm_exec_sql_text(sql_handle)) AS query_text
FROM sys.dm_exec_query_stats
ORDER BY [avg_cpu_cost] DESC

查詢缺失索引

SELECT 
    DatabaseName = DB_NAME(database_id)
    ,[Number Indexes Missing] = count(*) 
FROM sys.dm_db_missing_index_details
GROUP BY DB_NAME(database_id)
ORDER BY 2 DESC;

SELECT  TOP 10 
        [Total Cost]  = ROUND(avg_total_user_cost * avg_user_impact * (user_seeks + user_scans),0) 
        , avg_user_impact
        , TableName = statement
        , [EqualityUsage] = equality_columns 
        , [InequalityUsage] = inequality_columns
        , [Include Cloumns] = included_columns
FROM        sys.dm_db_missing_index_groups g 
INNER JOIN    sys.dm_db_missing_index_group_stats s 
       ON s.group_handle = g.index_group_handle 
INNER JOIN    sys.dm_db_missing_index_details d 
       ON d.index_handle = g.index_handle
ORDER BY [Total Cost] DESC;

定位問題后，新建非聚集索引

CREATE NONCLUSTERED INDEX IX_t_AccessControl_F4 ON dbo.t_AccessControl
(
    FObjectType
)include([FUserID], [FAccessType], [FAccessMask]) WITH( STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
GO

drop index IX_t_AccessControl_F4 on t_AccessControl

 CPU占用恢復正常

跟蹤模板和跟蹤文件下載，請使用SQL2008R2 版本：files.cnblogs.com/lyhabc/跟蹤模板和trace.rar

 

---

總結

從多次歷史經驗來看，如果CPU負載持續很高，但內存和IO都還好的話，這種情況下，首先想到的一定是索引問題，十有八九錯不了。

注意文章開頭貼出的客戶機器負載情況圖